Controllable Abstractive Sentence Summarization with Guiding Entities

Proceedings of the 28th International Conference on Computational Linguistics, pages 5668–5678 Barcelona, Spain (Online), December 8-13, 2020 5668 Controllable Abstractive Sentence Summarization with Guiding Entities Changmeng Zheng1,2, Yi Cai1,2∗, Guanjie Zhang1, Qing Li3 1School of Software Engineering, South China University of Technology, Guangzhou, China 2Key Laboratory of Big Data and Intelligent Robot (SCUT), Ministry of Education, China 3Department of Computing, The Hong Kong Polytechnic University, Hong Kong, China sethecharm@mail.scut.edu.cn,ycai@scut.edu.cn Abstract Entities are the major proportion and build up the topic of text summaries. Although existing text summarization models can produce promising results of automatic metrics, for example, ROUGE, it is difficult to guarantee that an entity is contained in generated summaries. In this paper, we propose a controllable abstractive sentence summarization model which generates summaries with guiding entities. Instead of generating summaries from left to right, we start with a selected entity, generate the left part first, then the right part of a complete summary. Compared to previous entity-based text summarization models, our method can ensure that entities appear in final output summaries rather than generating the complete sentence with implicit entity and article representations. Our model can also generate more novel entities with them incorporated into outputs directly. To evaluate the informativeness of the proposed model, we develop a fine-grained informativeness metrics in the relevance, extraness and omission perspectives. We conduct experiments in two widely-used sentence summarization datasets and experimental results show that our model outperforms the state-of-the-art methods in both automatic evaluation scores and informativeness metrics. 1 Introduction Automatic Text Summarization is a task of shortening and creating a concise version of a text that represents the most important or relevant information within the original text. The task can be divided into two subtasks: extractive and abstractive summarization. Recently, the research trend (Nallapati et al., 2016; Tan et al., 2017; Bansal and Passonneau, 2018; Kry´sci´nski et al., 2018) has shifted towards abstractive summarization since it can produce more natural and coherent texts than extractive methods. Sequence-to-sequence models with attention mechanism (Chopra et al., 2016; Nallapati et al., 2016) have attracted growing attention due to their state-of-the-art performance on abstractive summarization. However, most previous seq2seq models have two obvious limitations: The first limitation is that existing methods are difficult to identify salient entities and related events in the original articles (Cao et al., 2018). As a result, they are frequently found to be unfaithful to the input and lack of topic coherence. According to our observations, entities are important for summary generation in three points: (1) Summaries are mainly composed by entities extracted from the original texts. 77.1% and 71.5% of the noun phrases on the Gigaword and DUC-2004 datasets contain at least one entity, respectively. (2) Entities can reveal the main topics in an article (Newman et al., 2006), which can be utilized to generate the summaries of highly topic coherence. For example, the PER entity “KOBE-BRYANT” and the ORG entity “NBA” are mostly associated with topic “games” and “players”. (3) Entities (e.g. person names, locations and organizations) …

Content truncated to 500 words